This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: FAIL: gcc.c-torture/execute/950628-1.c execution, -O1
- To: Richard Kenner <kenner at vlsi1 dot ultra dot nyu dot edu>
- Subject: Re: FAIL: gcc.c-torture/execute/950628-1.c execution, -O1
- From: Richard Henderson <rth at redhat dot com>
- Date: Fri, 26 Jan 2001 12:05:20 -0800
- Cc: gcc at gcc dot gnu dot org
- References: <10101260135.AA10263@vlsi1.ultra.nyu.edu>
On Thu, Jan 25, 2001 at 08:54:37PM -0500, Richard Kenner wrote:
> If all we gain are file-level const variables, we
> gain extremely little for a lot of overhead.
I don't necessarily agree that there are so few file-level const
variables that it's not worth special casing them, however ...
> So I think we do need to find some way to make more effective use of /u.
... this has the ring of a solution in search of a problem.
Why don't we just get rid of /u memory entirely?
I think it's confusing to have "unchanging" memory change. We both
agree that we have to reuse "const" temporaries as otherwise stack
frames explode. You mentioned using alias sets for .got data; I
think we can use them everwhere we currently use /u. This has the
advantage of reusing stack slots as you'd prefer without calling
them "unchanging" to which I object.
r~